MATHEMATICAL LOGIC FOR APPLICATIONS

Transcription

1 MIKLÓS FERENCZI MIKLÓS SZŐTS MATHEMATICAL LOGIC FOR APPLICATIONS 2011 Abstract Contents Sponsorship Editorship Referee Technical editor Copyright ISBN

2 This book is recommended for those readers who have completed some introductory course in Logic. It can be used from the level MSc. It is recommended also to specialists who wish to apply Logic: software engineers, computer scientists, physicists, mathematicians, philosophers, linguists, etc. Our aim is to give a survey of Logic, from the abstract level to the applications, with an emphasis on the latter one. An extensive list of references is attached. As regards problems or proofs, for the lack of space, we refer the reader to the literature, in general. We do not go into the details of those areas of Logic which are bordering with some other discipline, e.g., formal languages, algorithm theory, database theory, logic design, artificial intelligence, etc. We hope that the book helps the reader to get a comprehensive impression on Logic and guide him or her towards selecting some specialization. Key words and phrases: Mathematical logic, Symbolic logic, Formal languages, Model theory, Proof theory, Non-classical logics, Algebraic logic, Logic programming, Complexity theory, Knowledge based systems, Authomated theorem proving, Logic in computer science, Program verification and specification.

3 Acknowledgement of support: Prepared within the framework of the project Scientific training (matemathics and physics) in technical and information science higher education Grant No. TÁMOP /2/A/KMR Prepared under the editorship of Budapest University of Technology and Economics, Mathematical Institute. Referee: Károly Varasdi Prepared for electronic publication by: Ágota Busai Title page design: Gergely László Csépány, Norbert Tóth ISBN: Copyright: , Miklós Ferenczi, Miklós Szőts, BME Terms of use of : This work can be reproduced, circulated, published and performed for non-commercial purposes without restriction by indicating the author s name, but it cannot be modified.

4

5 Contents 0 INTRODUCTION 3 1 ON THE CONCEPT OF LOGIC Syntax Basic concepts of semantics Basic concepts of proof theory On the connection of semantics and proof theory CLASSICAL LOGICS First-order logic Syntax Semantics On proof systems and on the connection of semantics and proof theory Logics related to first-order logic Propositional Logic Second order Logic Many-sorted logic On proof theory of first order logic Natural deduction Normal forms Reducing the satisfiability of first order sentences to propositional ones Resolution calculus Automatic theorem proving Topics from first-order model theory Characterizing structures, non-standard models Reduction of satisfiability of formula sets On non-standard analysis NON-CLASSICAL LOGICS Modal and multi-modal logics Temporal logic Intuitionistic logic Arrow logics Relation logic (RA) Logic of relation algebras Many-valued logic

6 2 MATHEMATICAL LOGIC FOR APPLICATIONS 3.6 Probability logics Probability logic and probability measures Connections with the probability theory LOGIC AND ALGEBRA Logic and Boolean algebras Algebraization of first-order logic LOGIC in COMPUTER SCIENCE Logic and Complexity theory Program verification and specification General introduction Formal theories Logic based software technologies Logic programming Programming with definite clauses On definability A general paradigm of logic programming Problems and trends KNOWLEDGE BASED SYSTEMS Non-monotonic reasoning The problem Autoepistemic logic Non-monotonic consequence relations Plausible inference Description Logic Bibliography 106 Index 116

7 0. INTRODUCTION 3 Chapter 0 INTRODUCTION 1. Logic as an applied science. The study of logic as a part of philosophy has been in existence since the earliest days of scientific thinking. Logic (or mathematical logic, from now logic) was developed in the 19th century by Gottlob Frege. Logic has been a device to research foundations of mathematics (based on results of Hilbert, Gödel, Church, Tarski), and main areas of Logic became full-fledged branches of Mathematics (model theory, proof theory, etc.). The elaboration of mathematical logic was an important part of the process called revolution of mathematics (at the beginning of the 20th century). Logic had an important effect on mathematics in the 20th century, for example, on algebraic logic, non-standard analysis, complexity theory, set theory. The general view of logic has changed significantly over the last 40 years or so. The advent of computers has led to very important real-word applications. To formalize a problem, to draw conclusions formally, to use formal methods have been important tasks. Logic started playing an important role in software engineering, programming, artificial intelligence (knowledge representation), database theory, linguistics, etc. Logic has become an interdisciplinary language of computer science. As with such applications, this has in turn led to extensive new areas of logic, e.g. logic programming, special non-classical logics, as temporal logic, or dynamic logic. Algorithms have been of great importance in logic. Logic has come to occupy a central position in the repertory of technical knowledge, and various types of logic started playing a key roles in the modelling of reasoning and in other special fields from law to medicine. All these developments assign a place to Applied Logic within the system of science as firm as that of applied mathematics. As an example for comparing the applications and developing theoretical foundations of logic let us see the case of artificial intelligence (AI for short). AI is an attempt to model human thought processes computationally. Many non-classical logics (such as temporal, dynamic, arrow logics) are investigated nowadays intensively because of their possible applications in AI. But many among these logics had been researched by mathematicians, philosophers and linguists before the appearance of AI only from a theoretical viewpoint and the results were applied in AI later (besides, new logics were also developed to meet the needs of AI). In many respects the tasks of the mathematician and the AI worker are quite similar. They are both concerned with the formalization of

8 4 MATHEMATICAL LOGIC FOR APPLICATIONS certain aspects of reasoning needed in everyday practice. Philosopher, mathematician and engineers all use the same logical techniques, i.e., formal languages, structures, proof systems, classical and non-classical logics, the difference between their approaches residing in where exactly they put the emphasis when applying the essentially same methods. 2. Classical and non-classical logics. Chapter 2 is devoted to classical first-order logic and to logics closely related to it, called classical logics. Classical first-order logic serves as a base for every logic, therefore it is considered as the most important logic. Its expressive power is quite strong (contrary to propositional logic, for example) and it has many nice properties, e.g. completeness, compactness, etc., (in contrast to second-order logic, for example). It is said to be the logic of mathematics, and its language is said to be the language of mathematics. The Reader is advised to understand the basic concepts of logic by studying classical first-order logic to prepare the study of other areas of logic. However, classical logics describe only static aspects of the modelled segment of the world. To develop a more comprehensive logical model multiple modalities are to be taken into consideration: - what is necessary and what is occasional, what is known and what is believed, what is obligatory and what is permitted, past, present, future, sources of information and their reliability, uncertainty and incompleteness of information among others. A wide variety of logics have been developed and put to use to model the aspects mentioned above (in artificial intelligence, computer science, linguistics, etc.). Such logics are called non-classical logics. We sketch some important ones among them in Chapter 3 without presenting the whole spectrum of these logics, which would be far beyond the scope of this book. 3. On the concept of logic. Since many kinds of special logics are used in applications, a general frame has been defined for logic (see Chapter 1), which is general but efficient enough to include many special logics and to preserve most of their nice properties. It is worth understanding logic at this general level for a couple of reasons. First, we need to distinguish the special and general features of the respective concrete logics anyway. Second, it often happens that researchers have to form their own logical model for a situation in real life. In this case they can specialize a general logic in a way suitable for the situation in question. The general theory of logic is a new, and quickly developing area inside logic. We note that there is a clear difference between a concrete logic (with fixed non-logical symbols ) and a class of concrete logics (only the logical symbols are fixed). The latter is a kind of generalization of the concrete ones, of course. Usually, by logic we understand a class of logics, but the reader should be careful, the term logic because the term is used also for a concrete logic. We must not confuse the different degrees of generalizations. 4. Areas of mathematics connected with logic. An important aspect of this study is the connection between Logic and the other areas of mathematics. There are areas of mathematics which are traditionally close to Logic. Such areas are: algebra, set theory, algorithm theory. For example, modern logic was defined originally in algebraic form (by Boole, De Morgan and Peirce). An efficient method in Algebra (in Logic) for problem solving is the following: translate the problem to Logic (to Algebra) and solve it in logical (in algebraic) form. The scientific framework of this kind of activity is

9 0. INTRODUCTION 5 the discipline called Algebraic Logic founded in the middle of the 20th century (by Tarski, Henkin, Sikorski, etc.). This area is treated in Chapter 4. There are areas in mathematics which originally seemed fairly remote from Logic but later important and surprising logical connections were discovered between them. For example, such an area is Analysis. In the sixties, Abraham Robinson worked out the exact interpretation of infinitesimals through a surprising application of the Compactness Theorem of First Order Logic. Many experts believe this theory to be a more natural model for differential and integral calculus than the traditional model, the more traditional ε δ method (besides analysis Robinson s idea was applied to other areas of Mathematics too, and this is called non-standard mathematics). This connection is discussed in Section We also sketch some connections between Logic and Probability theory (3.6.1). 5. The two levels of logics. Every logic has two important levels : the level of semantics and that of proof theory (or proof systems or syntax). For most logics these two levels (two approaches) are equivalent, in a sense. It is important to notice that both levels use the same formal language as a prerequisite. So every logic has three basic components: formal language, semantics and proof theory. We make some notices on these components, respectively. The concept of language is of great importance in any area of logics. When we model a situation in real life the first thing we choose is a suitable language more or less describing the situation in question. We note that today the theory of formal languages is an extensive, complex discipline and only a part of this theory is used in Logic directly. Logical semantics is the part of logic which is essentially based on the theory of infinite sets. In general, in the definitions of semantics, there are no algorithms. Nevertheless, it is extraordinarily important in many areas of applications. Semantics is a direct approach to the physical reality. Proof theory is the part of logic which is built on certain formal manipulations of given symbols. It is a generalization of a classical axiomatic method. The central concept of proof theory is the concept of a proof system (or calculus). Setting out from proof systems algorithms can be developed for searching proofs. These algorithms can then be implemented on computers. What is about the connection between these two approaches to Logic? The strength of a logical system depends on the degree of equivalence between the semantic and the proof-theoretical components of the logic (such result are grouped under the heading of completeness theorems for a particular logic). The two levels of logic together are said to be the double construction of logic. First-order logic is complete, therefore, the double construction of logic has repercussions with respect to the whole mathematics. In addition to strength of a logic there are famous limits of logics (also that of first-order logic): undecidability and incompleteness (see Church and Gödel s results). These limits have interesting practical, philosophical and methodological consequences for the whole science. Throughout the Chapters 1 and 2 (and partially in Chapter 3) we treat the main components of logic and their relationships systematically.

10 6 MATHEMATICAL LOGIC FOR APPLICATIONS 6. On the reading of the book. We suppose that the reader has some experience in Logic. This does not mean concrete prerequisites, but a kind of familiarity with this area. For example, the reader will understand Chapter 1, the general frame of Logic, more comprehensively if he/she knows concrete logics (in any case, the reader is urged to return once more to Chapter 1 after reading Chapters 2 and 3). Today Logic is a large area inside science. To give a more or less comprehensive overview of this huge domain, we were forced to be selective on the topics and the theoretical approaches we survey. As regards the proofs of the theorems and further examples connected with the subject, we must refer the reader to the literature. Some important references are: [34], [51], [11], [145], [23], [43], [96], [156], [5], [71]. 7. Acknowledgement. We say thanks to Gábor Sági for his useful advices and notices. Furthermore, we say thanks also to our students at the Technical University Budapest (to mathematicians and software engineers) for their valuable notes.

11 1. ON THE CONCEPT OF LOGIC 7 Chapter 1 ON THE CONCEPT OF LOGIC In this chapter we give general definitions pertaining to classical and nonclassical logics, which we specialize, detail and illustrate with examples later in the book. We present the general concepts concerning the main parts of logic: syntax, semantics, proof theory and their connection, respectively. 1.1 Syntax First, we sketch the general concept of syntax of logics (the language L). This kind of syntax does not differ essentially from the syntax used in symbol processing in Computer Science, or in a wider sense, musical notes in music or written choreography in dance. Syntax is given in terms of a set of symbols called alphabet and a set of syntactical rules. In this book we assume that the alphabet is countable (finite or infinite), with possible exceptions being explicitly mentioned when necessary. The alphabet of a logic consists of two kinds of symbols. One of them are the logical symbols, the other are the non-logical ones. There is associated with each non-logical symbol particular a natural number or 0, the arity corresponding to the symbol. The sequences of such arities have importance in logic, it is called the type of the language. A finite sequence of the alphabet which has a meaning (by definition) is called a formula. The set of formulas is denoted by F and is defined by the syntactical rules. Besides formulas, there are other finite sequences of the alphabet which have importance, namely terms. These sequences are used to build up formulas. The term expression covers both formulas and terms. Assume that in a language L there is given a set P of symbols (called atomic formulas) and another set Cn (called connectives) such that for every connective c Cn has a natural number (the rank) k. Then, the set F of formulas coincide with the smallest set satisfying the conditions (i) and (ii) below: (i) P F,

12 8 MATHEMATICAL LOGIC FOR APPLICATIONS (ii) c(α 1,... α k ) F, where α 1,... α k are arbitrary formulas in F, and the connective c has the rank k. The terminology logical language (or language, for short) is used at least in two contexts in the literature. The one is for a concrete logical language, the other for a class of such concrete languages (this latter is called general language). A general language is specified by the logical symbols used, while a concrete language is specified by the concrete non-logical symbols of the alphabet (for example, the operations and constants +,,, 0, 1 as non-logical symbols specify the concrete first-order language of real numbers as a specialization of the general language of first-order logic). Some remarks concerning syntax are: It is important to realize that the definition of a logical language, and also, almost the whole study of logic uses metalanguages. The definitions in this book use natural language as the metalanguage, as it is usual. Generally, a logical language can be defined by context-free formal grammar: the alphabet and the types of the expressions correspond to the set of terminal symbols and to the non-terminal symbols, respectively. A terminological remark: formulas in a logical language correspond to sentences in a formal grammar. The word sentence in a logical language means a special class of formulas, which cannot be specified by a formal grammar. Let us think of programming languages, where programs can be defined by a context-free grammar, but important aspects of syntactic correctness cannot be described by thereby. Syntax can be defined as an algebra too: formulas form the universe of the algebra and the operations correspond to the rules of syntax. This algebra is a word algebra with equality being the same as identity (two different formulas cannot be equal). With the sets of atomic formulas and logical connectives in the language we can associate the word algebra generated by the set of atomic formulas using the given logical connectives as algebraic operations, in the usual algebraic sense. We can use prefix (Polish notation), infix or postfix notations for the expressions of the language. For example, taking a binary operation symbol O and applying it to the expressions α and β, the notations Oαβ, αoβ and αβo are prefix, infix and postfix notations, respectively. Any of these notational conventions has advantages as well as disadvantages. For example, the infix notation can be read easily, but brackets and punctuation marks (commas and points) are needed in the alphabet, and also, various precedence rules must be specified. Infix and postfix notations are useful to manipulate formulas by computers; e.g. to evaluate expressions automatically. For automated processing the so-called parsing tree provides an adequate representation of expressions. There are two usual additional requirements concerning syntax:

13 1. ON THE CONCEPT OF LOGIC 9 The set of formulas should be a decidable subset composed from the alphabet (decidability of the syntax). Formulas should have the unique reading property. The unique reading property means that for every expression of the language there is only one way to construct it by the rules of syntax (that is, every expression has a unique parsing tree). Most logical languages have both properties. 1.2 Basic concepts of semantics First we introduce a general concept of logic, approaching this concept from the side of logical semantics. Let us assume that a language L is given. Semantics is defined by a class of models (interpretations) and a meaning function, which provides the meaning of an expression in a model. The following formal definition pertains to many well-known logics: 1.1 Definition (logic in the semantical sense) A logic in the semantical sense is a triple L S = F, M, m where F is a set of formulas in L, M is a class (class of models or structures), and m is a function (meaning function) on F M, where we assume that the range of m is a partially ordered set. Sometimes we shall denote the members of L S in this way: F L, M L, m L. 1.2 Definition The validity relation ( truth for a formula on a model ) is a relation defined on F M in terms of the meaning function m (notation: M α, where α F, M M) as follows: M α if and only if m(β, M) m(α, M) for every β F (1.1) where is the given partial ordering on the range of m and α is a fixed formula. If there is a maximal element in the range of m, then (1.1) means that M α if and only if m(α, M) is maximal (for two-valued logic M α if and only if m(α, M) is true). We note that it often happens that the validity relation is defined first, then the meaning function in terms of. Now, we list some definitions concerning logics above:

14 10 MATHEMATICAL LOGIC FOR APPLICATIONS 1.3 Definition A formula α is said to be universally valid if M α for every model M, where M M (notation: α). M is a model of a given set Σ of formulas if M α for every α Σ(notation: M Σ). α is a semantical consequence of a given set Σ of formulas if M α for every model M such that M Σ (notation: Σ α). The set of all semantical consequences of Σ is denoted by ConsΣ. A theory of a given class N of models is the set Γ of formulas such that α Γ if and only if M α for every M N (notation: T hn or T hm if N = {M}). A theory T hn is decidable if the set T hn is a decidable subset of F. In particular a logic is decidable if T hm is decidable, where M is the class of its models. If the truth values true and false are present in the range of the meaning function (they are denoted by t and f), then a formula α is called a sentence if m(α, M) {t, f} for every model M. 1.4 Definition A logic has the compactness property if the following is true for every fixed set Σ of formulas: if every finite set Σ (Σ Σ) has a model, then Σ also has a model. A logic has Deduction theorem if there is a formula scheme Φ(α, β) such that Σ {α} β is equivalent to Σ Φ(α, β) (an important special case when Φ(α, β) is α β). Some comments on these definitions: Models (or structures) represent a particular domain of real life in logic (in a sense). Notice that the symbol is used in three different senses (and the intended meaning is determined by the context): for validity relation: M α, for universal validity: α, for semantical consequence: Σ α. Compactness is an important property of a logic because it allows a kind of finitization. An equivalent version of compactness is the following: If Σ α, then Σ α for some finite subset Σ of Σ. Compactness and Deduction theorems together will provide a connection between the concepts of semantical consequence and universal validity. A version of the indirect inference rule is the following equivalence: Σ α if and only if Σ { α} has no model. Algebras can also be associated with models, but we skip the details here. Now we define another important concept of logic, that of regular logic. This is a stronger, but at the same time more specific concept than the concept of logic discussed above.

15 1. ON THE CONCEPT OF LOGIC 11 Let a language L and a model M be fixed. A connective is said to be derived in L if it can be defined in terms of the basic logical connectives of L. 1.5 Definition (regular logic) A logic L S is called a regular logic, if the following properties (i), (ii), (iii) are satisfied: (i) (principle of compositionality). Assume that in L with every logical connective c of rank k an operation C of rank k is associated in the range of m. Then, m(c(α 1,... α k ), M) = C(m(α 1, M),... m(α k, M)) must hold for arbitrary formulas α 1,... α k. (ii) Assume that is a binary derived connective in L and T is a derived constant (as special connective) with the meaning true. Then, are required. M α β if and only if m(α, M) = m(β, M), and M T β if and only if M β. (iii) ( substitution property) Assume that L contains a set Q of atomic formulas. Then, for an arbitrary formula α, containing the atomic formulas P 1,... P n α(p 1,... P n ) implies α(p 1 /β 1,... P n /β n ) must hold for arbitrary formulas β 1,... β n, where P 1 /β 1,... P n /β n denote the result of every occurrence of P i being substituted simultaneously with β i, i = 1,..., n. (i) means that m preserves syntactical operations, that is, from the algebraic viewpoint, m is a homomorphism from the word algebra of formulas to the algebra corresponding to the model. Compositionality ensures that the meanings of formulas in a fixed model constitute such an algebra which is similar to the word algebra (this is one of the foundations of the so-called algebraic semantics). In (ii), the operation is a weakening of the operation (biconditional); therefore, if the biconditional can be formulated in the language, can be replaced in (ii) by. For regular logics, it is possible to prove stronger (but still general) results than for those in Definition 1.1.

16 12 MATHEMATICAL LOGIC FOR APPLICATIONS 1.3 Basic concepts of proof theory First, we define a central concept of proof theory: the concept of proof system (or calculus). A proof system is defined by a set of axioms, a set of inference rules and the concept of proof. Now we sketch these concepts, respectively. Let us extend the language by an infinite sequence X 1, X 2,... of new variables (called formula variables). First, we define the concept of formula scheme by recursion. Formula schemes are obtained by applying finitely many times the following rules: (i) the formula variables X 1, X 2,... are formula schemes, (ii) if Φ 1, Φ 2,... are formula schemes and c is a k-ary logical connective in the language, then c(φ 1, Φ 2,..., Φ k ) is also a scheme. A formula α is an instance of a scheme Φ if α is obtained by substituting all the formula variables in Φ by given formulas. An axiom of a calculus (a logical axiom) is given as a formula scheme (but the term axiom is usually used for both the scheme and its instance). An inference rule is Φ 1, Φ 2,..., Φ n, Φ, where Φ 1, Φ 2,..., Φ n, Φ are formula schemes, Φ 1, Φ 2,..., Φ n are called the premises, Φ is called the conclusion. Another well-known notation for an inference rule is: Φ 1,Φ 2,...Φ n Φ. The next important component of proof systems is the concept of proof. There are several variants of this concept. We define an important one together with the concept of provability for the case of the so-called Hilbert style proof systems. This definition is very general and simple. Let us assume that a set of axioms and a set of inference rules are fixed (we skip the details). 1.6 Definition A formula α is provable (derivable) from a set Σ of formulas (notation: Σ α) if there is a finite sequence ϕ 1, ϕ 2,..., ϕ n (the proof for α) such that ϕ n = α and for every i n, (i) ϕ i Σ, or (ii) ϕ i is an instance of an axiom (scheme), or (iii) there are indices j 1, j 2,..., j k < i and an inference rule Φ 1, Φ 2,..., Φ k, Φ in the system such that ϕ, ϕ j1j j2,..., ϕ jk, ϕ i is an instance of this rule (i.e. the formulas ϕ j1j, ϕ j2,..., ϕ jk, ϕ i are instances of the schemes Φ 1, Φ 2,..., Φ k, Φ in this rule, respectively). There is an important additional requirement for proof systems: the set of axioms and the set of inference rules should be decidable.

17 1. ON THE CONCEPT OF LOGIC 13 The relation is called the provability relation. is a relation on P(F) F (where P(F ) is the power set of F). If the proof system (calculus) is denoted by C, then the provability relation corresponding to C is denoted by C (if misunderstanding is excluded we omit C from C ). With different proof systems C 1 and C 2 we associate different provability relations C1 and C2, but it is possible that the relations C1 and C2 coincide (this is true for every well-known calculus of first-order logic, for example). Notice that with the concept of a proof system and the set Σ of formulas above we can associate the classical axiomatic method. We can classify proof systems according to the way they are put to use. From this viewpoint there are two kinds of proof systems: deduction systems and refutation systems. For a deduction system we set out from premises to get the conclusions (e.g. Hilbert systems, natural deduction). For refutation systems we apply a kind of indirect reasoning: the premises and the negation of the desired conclusion are simultaneously assumed and we are going to force a contradiction in a sense to get the conclusion (e.g. resolution, analytic tableaux). Now we introduce the proof-theoretical concept of logic. Assume that a fixed proof system C is given. 1.7 Definition (logic in proof-theoretical sense) A logic is a pair L P = F, C where F is the set of formulas in L and C is the provability relation specified by the proof system C. Sometimes the dependence on L is denoted in the members of L P way: F L, L C. in this We list some definitions for proof theory (we omit C from C ). 1.8 Definition If Σ α holds, we say that α is a proof-theoretical consequence of Σ. The set {α : Σ α, α F} of all the proof-theoretical consequences of a fixed Σ is denoted by Ded Σ. A set Σ of formulas is a theory if Σ = Ded Σ (that is Σ is closed under the provability relation). A theory Σ is inconsistent if both Σ α and Σ α hold for some formula α. Otherwise, the theory Σ is said to be consistent. A theory Σ is complete if α Σ or α Σ holds for every formula α. A theory Σ is decidable if Σ is a decidable subset of F. A theory Σ is axiomatizable if Σ = Ded Σ for some recursive subset Σ of F. Now we list the most important general properties of a provability relation: (i) inclusion (reflexivity), that is β Σ implies Σ β, (ii) monotonicity, that is Σ α implies Σ {β} α

18 14 MATHEMATICAL LOGIC FOR APPLICATIONS (iii) cut, that is Σ ϕ and Λ {ϕ} α imply Σ Λ α, where α, β and ϕ are formulas, Σ and Λ are arbitrary sets of formulas. Finally, some words on the concept of automatic theorem proving (see Section 2.3.5). A proof system does not provide a decision procedure, that is, the provability relation is not a decidable relation. A proof system only provides a possibility for a proof. An old dream in mathematics is to generate proofs automatically. This dream got closer to reality in the age of computers. Algorithms have to be constructed from calculi from which a derivation of the required theorem is performed. Historically, resolution calculus was considered as a base for automatic theorem proving. Since then, the devices of automatic theorem proving have been multiplied. 1.4 On the connection of semantics and proof theory Now we turn to the connection between the two levels of logic, to the connection between semantics and proof theory. Let us consider a logic in semantic form and in syntactical form together, with the same set F of formulas: in this way, we obtain a more comprehensive notion of logic. 1.9 Definition A logic is the sequence L = F, M, m, C where the members of the sequence are the same as in the Definitions 1.1 and 1.7. To obtain stronger results (e.g., proving completeness), it is necessary to assume that the semantical part of L is a regular logic. We list some concepts concerning the connection between the consequence relation and a provability relation C Definition A proof system C (or the relation C, or the logic L) is strongly complete if Σ α implies Σ C α for every set Σ of the formulas and every formula α. If Σ C α implies Σ α for every Σ and α, then the proof system is said to be strongly sound. A proof system (or the relation C, or the logic L) is weakly complete if α implies C α for every formula α. In the opposite case, that is, if C α implies α, then the proof system is said to be weakly sound. Completeness theorems, i.e., theorems stating completeness together with soundness for a given logic, are basic theorems of logics. Most of the important logics have a kind of completeness property.

19 1. ON THE CONCEPT OF LOGIC 15 The strong completeness theorem is: Σ α if and only if Σ C α i.e. the semantical and the syntactical concepts of logical consequence are equivalent (ConsΣ and Ded Σ coincide). Remarks on completeness: The main idea of proof theory is to reproduce the semantical concept Σ α (or only the concept α), using only finite manipulations with formulas and avoiding the in-depth use of the theory of infinite sets included in the definition of Σ α. Strong completeness of a logic makes it possible for us to use such a short cut (weak completeness makes it available the reproduction for the case when Σ = ). Weak completeness and compactness imply strong completeness, as can be shown. Another important version of strong completeness is: A set Σ of formulas is consistent if and only if Σ has a model. This version is the base of the famous model method for proving the relative consistency of a system. Refutation systems impose a condition on a set Γ of formulas having no model. Using this condition and the fact that Σ α if and only if Γ = Σ { α} has no model, we can prove Σ α. The following problem is of central importance in logic and it is closely related to completeness and incompleteness: Is it possible to generate in a recursive way the formulas of T hk, where K is any fixed class of models, i.e., is there a recursive set Σ of formulas such that T hk = Ded Σ (1.2) There are two important special cases of (1.2), the cases when K = M and K = {M}, where M is a fixed model and M is the class of the possible models. If the logic has weak Completeness theorem, then in the case K = M the answer is affirmative for the problem (1.2). But in the case K = {M}, if T hk is strong enough (i.e. recursive relations can be defined in the theory), by Gödel Incompleteness Theorem, the formula does not exist in general, so the answer is negative for problem. Setting out from a logic in the semantical sense (Definition 1.1) we can speak of the (weak or strong) completeness of a logic without a concrete proof system.

20 16 MATHEMATICAL LOGIC FOR APPLICATIONS 1.11 Definition A logic (in the semantical sense) is complete (weakly or strongly) if there is a proof system C and a provability relation C such that supplementing the logic by C the resulting logic is complete (weakly or strongly). References to Chapter 1 are, for example: [30], [96], [11], [13].

21 2. CLASSICAL LOGICS 17 Chapter 2 CLASSICAL LOGICS In this chapter we mainly deal with first-order logic. The other logics treated in this chapter are closely related to the first-order one. By investigating them one can attain a better understanding about the limitations and the advantages of first-order logic. 2.1 First-order logic First-order logic (FOL) plays an exceptional role among logics. Any other classical logic either has less expressive power (e.g. propositional logic) or does not have the nice properties which first-order logic has. In a direct or an indirect way almost every logic can be reduced to first-order logic in a sense (this does not mean that other logics should be ignored). First-order logic is said to be the logic of classical mathematics. Though mathematics also uses propositional logic, second-order logic, etc. to a certain extent, the applications of these logics can be somehow simulated by first-order logic. The language L of firstorder logic can be considered also as a collection of basic mathematical notations. First-order logic is applied not only in mathematics but in almost every area of Science, where logic is applied at all Syntax The alphabet U of a first-order language L contains the connectives,,,, and with ranks 1, 2, 2, 2, 2, 1 and 1, respectively, the equality symbol =, a sequence x 1, x 2,... of individuum variables as logical symbols; furthermore a sequence f 1, f 2,... of function symbols (including the individuum constants) and a sequence P 1, P 2,... of relation symbols (including the propositional constants) as non-logical symbols. The numbers of the arguments of the function symbols and those of relation symbols are given by two sequences corresponding to those of function and relation symbols (with members of the sequences being natural numbers or 0, the two sequences being separated by a semicolon ;). The symbols,,,, and correspond to the words not, and, or, for every, for some, respectively. Defining a first-order language means to specify the concrete (finite or infinite) sequences of its function symbols and

22 18 MATHEMATICAL LOGIC FOR APPLICATIONS relation symbols (and also the sequences of the number of their argument). This double sequence is the type of the language. For example, the alphabet of the language L R of ordered reals contains the non-logical constants +,,, 0, 1 and (the signs of addition, multiplication, minus, zero, one, less than or equal, respectively). The type of L R is 2, 2, 1, 1, 0, 0; 2, where ; separates the arities of function symbols and that of relation symbols. Two remarks concerning the alphabet of L: Operations which can be expressed in terms of other operations can be omitted from the alphabet. For example, can be expressed in terms of and, therefore, can be omitted. Sometimes extra symbols are introduced for the truth values ( true and false ) into the logical language. In first-order languages individuum constants and propositional constants may form a separate category among non-logical symbols. Here we consider them as function symbols and relation symbols with 0 argument, respectively. We define the expressions of first-order languages: terms and formulas. 2.1 Definition Terms are obtained by finitely many applications of the following rules: (i) the individuum variables and the individuum constants are terms; (ii) if f is an n ary function symbol and t 1,... t n are terms, then ft 1,... t n is also a term. 2.2 Definition First-order formulas are finite sequences over the alphabet of L, obtained by finitely many applications of the following rules: (i) if P is a n ary relation and t 1,... t n are terms, then P t 1,... t n is a formula, (ii) t 1 = t 2 is a formula, where t 1, and t 2 are terms, (iii) if α and β are formulas, then α, αβ, αβ, αβ, αβ are formulas, (iv) if x is any individuum variable and α is a formula, then xα and xα are formulas. We note that first-order languages have the properties of unique readability and decidability. 2.3 Definition Formulas occuring in the definition of a given formula are called the subformulas of the given formula. A scope of a quantifier or in the formula xα or xα is the subformula α. A given occurrence of the variable x in a formula is bounded if this occurrence is in the scope of some quantifier, in the opposite case this occurrence is said to be free. A variable x is a free variable in a formula if it has a free occurence. A formula is said to be a sentence (to be a closed formula) if it has no free variable.

23 2. CLASSICAL LOGICS 19 The formulas in (i) and (ii) are called atomic formulas. The concept of the substitution of a free variable by another variable can also be defined, but, we skip the details. Definitions (2.2) and (2.3) and the examples above use prefix notations for the sake of brevity. But we use the usual infix notations when manipulating formulas by hand (assuming that the language is extended by precedence rules and brackets, well known from mathematical practice). Examples for prefix expressions in the language L R of reals: terms: +01, x1, x + y0, 1 y1, atomic formulas: xy1, 0 1 x, = 0xx, sentences: x y = xy1, x = x00, x = x0 = x 1 x1, formulas with free variable: y = xy1, = x0 = x 1 x1. the infix versions of expressions above: terms: 0 + 1, x 1, x (y + 0), y 1 1, atomic formulas: x y 1, 0 x 1, 0 x = x, sentences: x y(x y = 1), x(x 0 = 0), x( x = 0 x x 1 = 1), formulas with free variable: y(x y = 1), x = 0 x x 1 = Semantics 2.4 Definition A model (or structure) A of type of L is a sequence A = V A, P A 1, P A 2,..., f A 1, f A 2,... (2.1) where V A is a non-empty set (the universe of A), P1 A, P2 A,... are concrete relations on V A associated with the relation symbols P 1, P 2,... with arities given in L, f1 A, f2 A,... are concrete functions on V A associated with the function symbols f 1, f 2,... with arities given in L. Briefly put, a model is a set V A equipped with certain relations (with functions and individuum constants, in particular) on V A. The type of a model (structure) is that of the language used. The superindices A s in (2.1) are omitted if misunderstanding is excluded. The interpretations of the function-, constant- and relation symbols in L are defined in (2.1). The interpretations of terms on A can be defined as usual. However, the interpretation of variables is not determined by the model since individuum variables are expected to denote any element of the universe. A possible interpretation of the variables can be considered as a sequence q 1, q 2,..., q i,... with members from the universe V, corresponding to the sequence x 1, x 2..., x i,... of the individuum variables, respectively. Thus q 1, q 2,..., q i,... (q for short) is the function on the natural numbers such that q i V and x A i = q i, i = 1, 2,.... We are going to define the interpretation of a formula α on A with free variables. It will be defined as all the interpretations of the individuum variables under which α is true on A. This set of interpretations of the individuum variables will be called the truth set of α in A and it is defined by formula recursion as follows:

24 20 MATHEMATICAL LOGIC FOR APPLICATIONS 2.5 Definition (i) If P t 1,... t n is an atomic formula, including the individuum variables x i1,... x ik, where P is P 1 or P 2 or P 3 etc., then the truth set [P t 1,... t n ] of P t 1,... t n in A is the set { q t A 1 (q),..., t A n (q) P A} (2.2) in particular the truth set [P (x i1,... x ik )] is { q qi1... q in P A}. (2.3) Also, if P t 1,... t n is t 1 = t 2, then [t 1 = t 2 ] is {q t 1 (q) = t 2 (q)}. (ii) If [α] and [β] are defined, then let the truth sets [ α], [α β], [α β], [α β], [α β] be V ω [α], [α] [β], [α] [β], [ α] [β], [ α] [ β] [α] [β], respectively. (iii) If [α] is defined, let [ x i α] = { q q i v [α] for some v V } and [ x i α] = { q q i v [α] for every v V } where q i v is obtained from q by substituting the ith member of q i with v. We define the concepts of the meaning function and validity relation in terms of the concept of truth set: 2.6 Definition The value of the meaning function m for the formula α and model A is the truth set [α] of α in A where the partial ordering on V ω is set inclusion. It is easy to check that A α if and only if [α] = V ω. The truth values true and false are represented by V ω and. In particular, if α is a sentence (a closed formula), α is true on A if and only if [α] = V ω. α is called to be true for an evaluation q of the individuum variables (in notation A α [q] ) if and only if q [α]. One of the main purposes in logic is to find the true propositions on a given model, that is to find the theory of the model. For example, let R = R, + R, R, 0 R, 1 R R is the structure of the ordered real numbers, where R is the set of real numbers, and the operations, constants and the relation R are the usual ones. The type of R is the same as that of L R. The theory T hr of the reals consists of all the true propositions on R, formulated in terms of L R.

25 2. CLASSICAL LOGICS 21 Some comments on these definitions: To understand the intuition behind the concept of a truth set, see the following example. Let us consider the formula x 3 x 4. The truth set of this formula is: where is the interpretation of. {q q V ω, q 3 q 4, q 3, q 4 V } (2.4) One can see that only the third and fourth members of the q s play a role in (2.4). Therefore, we can rewrite (2.4) in this way: { q 3, q 4 q 3 q 4, q 3, q 4 V }. In general, since every formula has only finitely many free individuum variables, a truth set depends on only finitely many members of the sequence of the individuum variables. Nevertheless, for the sake of uniformity, we assume that formally the relations corresponding to formulas, i.e. the truth sets are infinite dimensional. From a geometrical point of view this means that a truth set corresponding to a formula with n free variables can be seen as an infinite dimensional cylinder set based to an n-dimensional set. Notice that the truth of a proposition is encoded in our Definition(2.5) as the relation, in a sense: A α [q] if and only if q [α]. This corresponds to the general definition of relations: a relation R is called true in a point q if and only if q R. We note that there is another, more traditional way to define the relation A α, without the concept of truth sets. Notice that (iii) in Definition (2.5) reflects exactly the intended meaning of quantifiers for every element of the universe or for some elements of the universe. Infinite unions or intersections also can be used to define quantifiers. We treat first-order logic with equality. This means that equality = is introduced in the language as a logical symbol and by definition, it denotes the identity relation in every model. Notice that if the quantifiers were omitted from the language, the expressive power of the logic would be much weaker (we would obtain a kind of propositional logic) On proof systems and on the connection of semantics and proof theory Some important positive results for first-order logic:

26 22 MATHEMATICAL LOGIC FOR APPLICATIONS 2.7 Theorem First-order logic is weakly complete and sound, so α if and only if α for some provability relation. In other words the theory of first-order logic is axiomatizable. 2.8 Theorem First-order logic is compact. A consequence of the theorems above is: First-order logic is strongly complete. The so-called Hilbert proof system, for example, is a strongly complete proof system for first-order logic. But there are many other well-known strongly complete proof systems for first-order logic: deduction systems (natural deduction, Gentzen s sequent calculus, etc.) and refutation systems (resolution, analytic tableaux), too. The axioms and inference rules of the Hilbert proof system: 2.9 Definition Axioms for first-order Hilbert system (Hilbert calculus) (i) α (β α). (ii) (α (β δ)) ((α β) (α δ)). (iii) ( α β) (β α). (iv) x(α β) (α xβ), where x is not free in α. (v) xα(... x...) α(... x/t...), where... x... denotes that x is a free variable in α, and t is such a term that the free variables occuring in t remain free after applying the substitution x/t. (vi) x = x, (vii) x = y t(... x...) = t(... x/y...). (viii) x = y α(... x...) = α(... x/y...). The axioms (vi)-(viii) are called the axioms of equality Definition The inference rules are: Φ Ψ, Φ, Ψ ( modus ponens), Φ, xφ ( generalization). We must not confuse the relation and the connective but of course the deduction theorem establishes an important connection between them. The deduction theorem says that σ α if and only if σ α, where σ and α are sentences. Among others, as such, the investigation of the consequence relation σ α can be reduced to the investigation of the special relation σ α. This result can be generalized from σ to a finite set Σ of closed formulas. There are famous limitations of first-order logic: First-order logic is not decidable (by Church theorem). This means that the theory of the deducible sentences (i.e. T hm, where M is the class of first-order models) is not decidable.

27 2. CLASSICAL LOGICS 23 If a first-order theory is strong enough (i.e. recursive relations can be interpreted in the theory), then the theory is not axiomatizable, by Gödel s first Incompleteness Theorem; as a consequence, not decidable either (undecidability can be derived from Church theorem). Another version of incompleteness is the following: if an axiomatizable, consistent first-order theory is strong enough, then it is not complete. In general, a model M is not determined by its theory T hm (by Löwenheim Skolem theorem, see Section 3.4.1). If a theory is strong enough, then its inconsistency cannot be proven inside this theory (by Gödel s second Incompleteness Theorem). References to Section 2.1 are, for example: [51], [23], [25]. 2.2 Logics related to first-order logic In this section we are concerned with propositional logic as a restriction, and with second-order logic as an extension of first-order logic, respectively. Propositional logic is also called as 0th order logic (we do not discuss nth order logics, in general). Finally, we survey many-sorted logic which is a version of first-order logic (it is equivalent to that, in a sense) Propositional Logic Classical propositional logic is a base for every logic. It has many nice properties but its expressive power is not very strong. Nevertheless, there are many applications of this logic. It plays a central role in logical design (e.g. at designing electrical circuits) or at the foundations of probability theory (see the concept of algebras of events, Section 4.2), among others. We introduce the language and the semantics of propositional logic independently of first-order logic. The alphabet of the language contains the logical connectives,,,, and the infinite sequence P 1, P 2,... of propositional symbols as non-logical symbols Definition Formulas are obtained by finitely many applications of the rules below: (i) the propositional constants are formulas, (ii) if α and β are formulas, then α, αβ, αβ, αβ, αβ are also formulas. Let P denote the set of propositional constants.